Artificial intelligence is increasingly involved in decisions that affect people’s lives, from credit approvals and healthcare recommendations to content ranking, fraud detection, and hiring support systems.

As AI becomes more capable, a fundamental question remains: How do we know when to trust it?

A model can make the right prediction while being overconfident. It can provide reliable predictions overall while being less reliable for specific demographic groups. And even when fairness safeguards exist, organisations often lack practical tools to continuously monitor whether those safeguards are working in production.

From the ongoing AI wave, we can all collectively agree that for AI to be trustworthy, accuracy alone is no longer enough. It needs reliability. That is why our Dolby researchers are building AI that knows its limits, treats people fairly, and earns trust. From graph-based recommendation systems and social networks to credit risk assessment and high-stakes decision support, Dolby’s approach lays the groundwork for AI systems that are not only powerful, but also more accountable, equitable, and trustworthy.

Recent research is tackling these challenges through a new generation of methods focused on trustworthy AI. Rather than simply improving prediction performance, these efforts seek to answer three critical questions:

How confident should we be in an AI prediction?
Are those confidence guarantees distributed fairly across different groups?
How can people continuously audit and govern AI systems in real-world environments?

For more in-depth research, explore the papers below:

The result is a complementary body of work that spans uncertainty quantification, fairness-aware machine learning, and responsible AI governance. Together, these advances help transform AI from a prediction engine into a system that can communicate uncertainty, evaluate fairness, and provide transparency for continuous oversight.

From Accurate Predictions to Trust: Dolby’s Journey Toward Responsible AI

{{ pageEducationalSlot.headingContent[index] }}